Microsoft Researchers Show AI Agents Can Be Backdoored via Neuron Edits

# Microsoft Researchers Show AI Agents Can Be Backdoored via Neuron Edits

https://gcdradar.vercel.app/incidents/microsoft-researchers-show-ai-agents-can-be-backdoored-via-n · AIR-2026-1896 · live

- Lane: Research
- Severity: 1 (Research / proof of concept): Research proof of concept demonstrated by researchers with no known real-world exploitation.
- Fix status: Fix status unknown
- First reported: 2026-11-11
- AI role: Attack on AI
- Products: AI agents

## Summary

Microsoft researchers demonstrated that AI models can be backdoored by editing fewer than 10 rarely used neurons without retraining. This triggers malicious commands that connected AI agents can execute with their permissions.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-11-11: Microsoft Researchers Show AI Agents Can Be Backdoored by Editing Fewer Than 10 Neurons (https://thehackernews.com/2026/11/microsoft-researchers-show-ai-agents.html)

## Sources

- 2026-11-11 The Hacker News: Microsoft Researchers Show AI Agents Can Be Backdoored by Editing Fewer Than 10 Neurons https://thehackernews.com/2026/11/microsoft-researchers-show-ai-agents.html
- 2026-11-11 Help Net Security: Microsoft researchers demonstrate hidden backdoor technique for AI agents https://helpnetsecurity.com/2026/11/11/microsoft-hidden-ai-backdoors-lazarus
- 2026-11-13 arXiv: Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents https://arxiv.org/abs/2611.15118

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.